This paper introduces a hybrid model for Automatic Modulation Recognition (AMR) that enhances classification accuracy of digital communication signals across various signal-to-noise ratios (SNR). The proposed model integrates a Deep Learning (DL) layer featuring a Convolutional Neural Network (CNN) for initial classification with a subsequent refinement layer using the K-Means Clustering algorithm. The process begins with the continuous wavelet transform (CWT) of the received signal to generate a magnitude scalogram, which is used as an input image to the DL Layer. Due to the inherent loss of phase information in generating the magnitude scalogram, the system activates the K-means layer when the initial classification of the signal is a higher-order Phase Shift Keying (PSK) modulation scheme. This subsequent layer refines the classification by evaluating the In-phase and Quadrature (IQ) data and determines the modulation order of the PSK signal by use of the K-means algorithm. The MATLAB simulations showed that incorporating the K-means layer resulted in a 15% improvement in recognition accuracy, ultimately achieving over 95% accuracy for SNR levels of 15 dB and higher.